Search papers, labs, and topics across Lattice.
The paper introduces CoopDiff, a diffusion-based cooperative perception framework designed to enhance robustness against corruptions in multi-agent perception systems. CoopDiff employs a teacher-student architecture where a quality-aware teacher generates clean supervision features using a diffusion denoiser, and a dual-branch diffusion student reconstructs these features while adaptively integrating ego and cooperative information via an ego-guided cross-attention mechanism. Experiments on OPV2Vn and DAIR-V2Xn benchmarks, featuring diverse corruptions, demonstrate that CoopDiff outperforms existing methods and provides a tunable trade-off between accuracy and efficiency.
Diffusion models can significantly enhance the robustness of cooperative perception systems against real-world corruptions, achieving state-of-the-art performance by leveraging denoising properties.
Cooperative perception lets agents share information to expand coverage and improve scene understanding. However, in real-world scenarios, diverse and unpredictable corruptions undermine its robustness and generalization. To address these challenges, we introduce CoopDiff, a diffusion-based cooperative perception framework that mitigates corruptions via a denoising mechanism. CoopDiff adopts a teacher-student paradigm: the Quality-Aware Teacher performs voxel-level early fusion with Quality of Interest weighting and semantic guidance, then produces clean supervision features via a diffusion denoiser. The Dual-Branch Diffusion Student first separates ego and cooperative streams in encoding to reconstruct the teacher's clean targets. And then, an Ego-Guided Cross-Attention mechanism facilitates balanced decoding under degradation by adaptively integrating ego and cooperative features. We evaluate CoopDiff on two constructed multi-degradation benchmarks, OPV2Vn and DAIR-V2Xn, each incorporating six corruption types, including environmental and sensor-level distortions. Benefiting from the inherent denoising properties of diffusion, CoopDiff consistently outperforms prior methods across all degradation types and lowers the relative corruption error. Furthermore, it offers a tunable balance between precision and inference efficiency.
